Determining the Daytime Earth Radiative Flux from National Institute of Standards and Technology Advanced Radiometer (NISTAR) Measurements” by Wenying Su Et Al

Home , Radiative flux

Generated using version 3.2 of the oﬃcial AMS LATEX template

1 Determining the Daytime Earth Radiative Flux from National

2 Institute of Standards and Technology Advanced Radiometer

3 (NISTAR) Measurements

1 2 2 2 4 Wenying Su , Patrick Minnis , Lusheng Liang , David P. Duda ,

Konstantin Khlopenkov2, Mandana M. Thieman2, Yinan Yu3, Allan Smith3,

Steven Lorentz3, Daniel Feldman4, Francisco P. J. Valero5

1Science Directorate, NASA Langley Research Center, Hampton, Virginia

2Science Systems & Applications, Inc., Hampton, Virginia

3L-1 Standards and Technology, Inc., New Windsor, Maryland

4Lawrence Berkeley National Laboratory, MS 84R0171, Berkeley, California

5Scripps Institute of Oceanography, University of California, San Diego, CA, USA

1 5 ABSTRACT

6 The National Institute of Standards and Technology Advanced Radiometer (NISTAR) on-

7 board Deep Space Climate Observatory (DSCOVR) provides continuous full disc global

8 broadband irradiance measurements over most of the sunlit side of the Earth. The three ac-

9 tive cavity radiometers measures the total radiant energy from the sun-lit side of the Earth in

10 shortwave (SW, 0.2-4 µm), total (0.4-100 µm), and near-infrared (NIR, 0.7-4 µm) channels.

11 The Level 1 NISTAR dataset provides the ﬁltered radiances (the ratio between irradiance

12 and solid angle). To determine the daytime top-of-atmosphere (TOA) shortwave and long-

13 wave radiative fluxes, the NISTAR measured shortwave radiances must be unfiltered first.

14 An unﬁltering algorithm was developed for the NISTAR SW and NIR channels using a spec-

15 tral radiance data base calculated for typical Earth scenes. The resulting unﬁltered NISTAR

16 radiances are then converted to full disk daytime SW and LW ﬂux, by accounting for the

17 anisotropic characteristics of the Earth-reﬂected and emitted radiances. The anisotropy fac-

18 tors are determined using scene identiﬁcations determined from multiple low Earth orbit and

19 geostationary satellites and the angular distribution models (ADMs) developed using data

20 collected by the Clouds and the Earth’s Radiant Energy System (CERES). Global annual

21 daytime mean SW ﬂuxes from NISTAR are about 6% greater than those from CERES, and

22 both show strong diurnal variations with daily maximum-minimum diﬀerences as great as

−2 23 20 Wm depending on the conditions of the sunlit portion of the Earth. They are also

24 highly correlated, having correlation coeﬃcients of 0.89, indicating that they both capture

25 the diurnal variation. Global annual daytime mean LW ﬂuxes from NISTAR are 3% greater

26 than those from CERES, but the correlation between them is only about 0.38.

1 27 1. Introduction

28 The Earth’s climate is determined by the amount and distribution of the incoming so-

29 lar radiation absorbed and the outgoing longwave radiation (OLR) emitted by the Earth.

30 Satellite observations of Earth Radiation Budget (ERB) provide critical information needed

31 to better understand the driving mechanisms of climate change; the ERB has been moni-

32 tored from space since the early satellite missions of the late 1950s and the 1960s (House

33 et al. 1986). Currently, the Clouds and the Earth’s Radiant Energy System (CERES) in-

34 struments (Wielicki et al. 1996; Loeb et al. 2016) have been providing continuous global

35 top-of-atmosphere (TOA) reﬂected shortwave radiation and OLR since 2000. CERES data

36 have been crucial to advance our understanding of the Earth’s energy balance (e.g., Tren-

37 berth et al. 2009; Kato et al. 2011; Loeb et al. 2012; Stephens et al. 2012), aerosol direct

38 radiative eﬀects (e.g., Satheesh and Ramanathan 2000; Zhang et al. 2005; Loeb and Manalo-

39 Smith 2005; Su et al. 2013), aerosol-cloud interactions (e.g., Loeb and Schuster 2008; Quaas

40 et al. 2008; Su et al. 2010b), and to evaluate global general circulation models (e.g., Pincus

41 et al. 2008; Su et al. 2010a; Wang and Su 2013; Wild et al. 2013).

42 The Earth’s radiative ﬂux data record is augmented by the launch of the Deep Space

43 Climate Observatory (DSCOVR) on February 11, 2015. DSCOVR is designed to continu-

44 ously monitor the sunlit side of the Earth, being the ﬁrst Earth-observing satellite at the

45 Lagrange-1 (L1) point, ∼1.5 million km from Earth, where it orbits the Sun at the same

46 rate as the Earth (see Figure 1a). DSCOVR is in an elliptical Lissajous orbit around the

47 L1 point and is not positioned exactly on the Earth-sun line, therefore only about 92∼97%

48 of the sunlit Earth is visible to DSCOVR. As illustrated in Figure 1b, the daytime portion

49 (Ah) is not visible to the DSCOVR. Strictly speaking, the measurements from DSCOVR

50 are not truly ‘global’ daytime measurements. However, for simplicity we refer to them as

51 global daytime measurements. Onboard DSCOVR, the National Institute of Standards and

52 Technology Advanced Radiometer (NISTAR) provides continuous full disc global broadband

53 irradiance measurements over most of the sunlit side of the Earth (viewing the sunlit side

2 54 of the Earth as one pixel). Besides NISTAR, DSCOVR also carries the Earth Polychromatic

55 Imaging Camera (EPIC) which provides 2048 by 2048 pixel imagery 10 to 22 times per day

56 in 10 spectral bands from 317 to 780 nm. On June 8, 2015, more than 100 days after launch,

57 DSCOVR started orbiting around the L1 point.

58 The NISTAR instrument was designed to measure the global daytime shortwave (SW)

59 and longwave (LW) radiative ﬂuxes. The original objective of NISTAR was to monitor

60 the energy from the sunlit side of the Earth continuously, and to understand the eﬀects of

61 weather systems and clouds on the daytime energy. However, one limitation of NISTAR is

62 its relatively low signal-to-noise ratios, which necessitates averaging signiﬁcant time periods

63 to adequately reduce the instrument noise levels. This constrains the temporal resolution

64 of meaningful results to about 4 hours, thus prevent us from “continuously” monitoring the

65 sunlit side of the Earth. Nevertheless, NISTAR measurements can still be useful for assessing

66 the hourly ﬂuxes produced by combining the observations from multiple low-Earth orbit and

67 geostationary satellites (Doelling et al. 2013) and for model evaluation using the spectral

68 ratio information (Carlson et al. 2019). NISTAR measures an irradiance at the L1 point at

◦ ◦ 69 a small relative azimuth angle, φo, which varies from 4 to 15 , as shown in Figure 1a. As

70 such, the radiation it measures comes from the near-backscatter position, which is diﬀerent

71 from that seen at other satellite positions as indicated in Figure 1a by the varying arrow

72 lengths corresponding to scattering angles, Θ1 − Θ3. Other types of Earth-orbiting satellites

73 view a given spot on the Earth from various scattering angles that vary as a function of local

74 time (e.g., geostationary) or overpass time (e.g., Sun-synchronous). When averaged over the

75 globe, the uncertainties in the anisotropy corrections are mitigated by compensation. That

76 is, any small biases at particular angles are balanced by observations taken at other angles.

77 In contrast, instruments on DSCOVR view every spot on the Earth from a single scattering

78 angle that varies slowly within a small range over the course of the Lissajous orbit. Thus, the

79 correction for anisotropy is critical. The biases in the anisotropy correction for the DSCOVR

80 scattering angle are mitigated and potentially minimized by the wide range of diﬀerent scene

3 81 types viewed in a given NISTAR measurement (Su et al. 2018).

82 Su et al. (2018) described the methodology to derive the global mean daytime shortwave

83 (SW) anisotropic factors by using the CERES angular distribution models (ADMs) and a

84 cloud property composite based on lower Earth orbiting satellite imager retrievals. These

85 SW anisotropic factors were applied to EPIC broadband SW radiances, that were estimated

86 from EPIC narrowband observations based upon narrowband-to-broadband regressions, to

87 derive the global daytime SW ﬂuxes. Daily mean EPIC and CERES SW ﬂuxes calculated

88 using concurrent hours agree with each other to within 2%. They concluded that the SW

89 ﬂux agreement is within the calibration and algorithm uncertainties, which indicates that

90 the method developed to calculate the global anisotropic factors from the CERES ADMs

91 is robust and that the CERES ADMs accurately account for the Earth’s anisotropy in the

92 near-backscatter direction.

93 In this paper, the same global daytime mean anisotropic factors developed by Su et al.

94 (2018) are applied to the NISTAR measurements to derive the global daytime mean SW

95 and longwave (LW) ﬂuxes. The NISTAR data and the unﬁltering algorithms developed for

96 the NISTAR shortwave and near-infrared channels are detailed in section 2. The data and

97 methodology used to derive the global daytime mean anisotropic factors are presented in

98 section 3. Hourly daytime SW and LW ﬂuxes calculated from NISTAR measurements and

99 comparisons with the CERES Synoptic ﬂux products (SYN1deg, Doelling et al. 2013) are

100 detailed in section 4, followed by conclusions and discussions in section 5.

101 2. NISTAR observation

102 The NISTAR instrument measures Earth irradiance data for an entire hemisphere using

103 cavity electrical substitution radiometers (ESRs) and ﬁlters covering three channels: short-

104 wave (SW, 0.2-4.0 µm), near-infrared (NIR, 0.7-4.0 µm), and total (0.2-100 µm). Each

105 channel has a dedicated ESR, that by itself is sensitive to radiation from 0.2-100 µm. For

4 106 the NIR and SW channels, ﬁlters are positioned in front of each ESR to limit the incident

107 radiation to spectral bands. The ﬁlters reside in a ﬁlter wheel that, during normal operation,

108 conﬁgures each ESR to measure contemporaneously in a diﬀerent band. Additionally, each

109 ESR has a shutter that modulates the Earth signal by cycling between open and closed states

110 continually with a 50% duty cycle and a period of 4 minutes. The modulation is necessary

111 as the ESRs only measure changes in the incident optical power and, being thermal detec-

112 tors, they have large oﬀsets (background signals) which drift over relatively short time frames

113 (hours) but not signiﬁcantly over a shutter cycle. Demodulating the resulting signal removes

114 those oﬀsets and the associated drifts/noise. What remains is a much more stable shutter

115 modulated background that is measured during periodic views of dark space and subsequently

116 subtracted from the signal. The shutter modulated background is largest for the total channel

117 and much smaller for the SW and NIR channels.

118 The NISTAR calibrated Level 1B data products are derived from pre-launch system level

119 optical calibration and on-orbit oﬀset measurements. The former involved optical response

120 measurements of each active cavity radiometer without a ﬁlter in place using a narrow band

121 calibration source whose irradiance was measured with a NIST calibrated reference detector.

122 Those measurements establish the irradiance responsivity of each spectrally ﬂat broadband

123 radiometer. Additionally, measurements of the transmittance of the SW and NIR ﬁlters

124 were made. This was done at NIST prior to installation into the NISTAR ﬁlter wheel

125 at wavelengths ranging from 200 nm to approximately 18 micrometers. Further, system-

126 level ﬁlter transmittance measurements at discrete visible and near-infrared wavelengths were

127 made using the external light source and the NISTAR photodiode channel as a detector. The

128 two transmittance measurements agreed to within a few tenths of a percent. Radiometric

129 offsets are measured on-orbit monthly when NISTAR briefly views dark space. The offset

130 measurement uncertainty is determined by the instrument noise level and the relatively short

132 NISTAR Level 1B radiometric products are derived by ﬁrst subtracting the oﬀsets from

5 133 Earth-view measurements and then dividing by the laboratory measured responsivity. The

134 result is irradiance measured at the instrument aperture. Radiance (I) is then calculated

135 from the irradiance data and the solid angle (Θ) determined from the DSCOVR-to-Earth

136 distance and the Earth dimensions. When averaging over a 4-hour period, the NISTAR

137 total and SW channel uncertainties (k=1) are 1.5% and 2.1%, respectively. As the LW is

138 derived from the diﬀerence between the total and unﬁltered-SW channels, it contains noise

−2 139 contributions from both. The LW uncertainty is about 3.3% (8 Wm ) given that the daytime

−2 −2 140 mean LW and SW ﬂuxes are approximately 210 Wm and 240 Wm , respectively, and that

141 the uncertainties between the Total and SW channels are largely uncorrelated.

142 As mentioned before, Filters are placed in front of the radiometers to measure the energies

143 from the SW and NIR portions of the spectrum. Since no corrections for the impact of

144 ﬁlter transmission were applied to the NISTAR L1B data, the SW and NIR radiances from

145 NISTAR must ﬁrst be unﬁltered before they can be used to derive daytime Earth’s radiative

146 ﬂux. Here we follow the algorithm developed by (Loeb et al. 2001) to convert measured

147 NISTAR ﬁltered radiances to unﬁltered radiances.

148 Unﬁltered SW and NIR radiances are deﬁned as follows:

Z λ2 band Iu = Iλdλ, (1) λ1

−2 −1 149 where ‘band’ represent either SW or NIR, λ(µm) is the wavelength, and Iλ (Wm sr

−1 150 µm ) is the spectral SW radiance. The ﬁltered radiance is the radiation that passes through

151 the spectral ﬁlter and is measured by the detector:

Z λ2 band band If = Sλ Iλdλ, (2) λ1

band 152 where Sλ is the spectral transmission function. Figure 2 shows the NISTAR SW and NIR

153 spectral transmission functions. These functions are determined from ground testing done

154 in 1999 and 2010 at the National Institute of Standards and Technology (NIST).

155 The spectral radiance database is calculated using high-spectral-resolution radiative trans-

156 fer model (Kato et al. 2002). Unﬁltered radiances are determined by integrating spectral

6 157 radiances over the appropriate wavelength intervals using Gaussian quadrature. Similarly,

158 ﬁltered radiances are computed by integrating over the product of spectral radiance and

159 spectral transmission function. The regression coeﬃcients are derived at 480 angles: 6 so-

160 lar zenith angles (0.0, 29.0, 41.4, 60.0, 75.5, 85.0 degrees), 8 viewing zenith angles (0, 12,

161 24, 36, 48, 60, 72, 84 degrees), and 10 relative azimuth angles (0 to 180, at every 20 de-

162 grees). For angles between those given above, the regression coeﬃcients are derived by linear

163 interpolation.

164 The database includes spectral radiances calculated over ocean, land/desert, snow/ice

165 surfaces for clear and cloudy conditions. Table 1 summarizes the number of each variable

166 that are included in the database, there are a total of 722 clear-sky cases and a total of 1519

167 cloudy-sky cases for each Sun-viewing geometry. This is a much larger database comparing

168 with that used by Loeb et al. (2001).

169 For CERES unfiltering, regression coefficients between filtered and unfiltered radiances

170 were derived as functions of scene type and Sun-viewing geometry (Loeb et al. 2001). Given

171 that NISTAR views the Earth as a single pixel, a mix of scenes and many Sun-viewing

172 geometries are observed at the same time. The method used for CERES is not feasible for

173 unﬁltering NISTAR observation. We instead investigated the feasibility of using the ratio,

174 κ, between filtered and unfiltered radiances for unfiltering the NISTAR observations. Table 2

175 lists the mean and the standard deviations of the ratios at diﬀerent solar zenith angles. The

176 ratios for the SW band are extremely stable, varying less than 0.3% among the scenes and

177 Sun-viewing geometries considered (the smallest ratio, 0.8659, occurs for clear ocean under

178 overhead sun and the largest ratio, 0.8694, occurs for clear/cloudy land under overhead

179 sun). As the ratio is not sensitive to the scene type and the Sun-viewing geometry, the SW

180 unﬁltering for NISTAR can be accomplished by: Isw Isw = f , (3) u κsw

sw 181 Here If is the ﬁltered radiances directly from the NISTAR L1B data. As the NISTAR view

182 always contains clouds, we choose to use the mean ratios of the cloudy ocean and land cases

7 183 in Table 2, which is 0.8690 for the SW band. The estimated uncertainty of using this single

184 ratio for unﬁltering the SW band is less than 0.3%.

185 On the other hand, the variability in the ratios of the NIR band can be as large as

186 6%. Fortunately, the large variability only occurs between clear ocean and clear land. As

187 mentioned earlier, NISTAR view always contains clouds and the mean ratios of the cloudy

188 ocean and land cases, which is 0.8583, is used to unﬁlter the NISTAR NIR observations.

189 This mean ratio can diﬀer with the individual ratios for diﬀerent solar zenith angles under

190 cloudy conditions by about 1∼2%. The mean ratio of the NIR bands is used to convert the

191 ﬁltered radiances to unﬁltered radiances: Inir Inir = f . (4) u κnir

192 In this paper, the measurements from NISTAR NIR channel are not used. The unﬁltering

193 of NIR channel is reported here for readers who intend to use this channel.

194 As there is no ﬁlter placed in front of the total channel, the radiance from the total

195 channel does not need to be unﬁltered. The LW (4-100 µm) radiance can be derived by

196 subtracting the unﬁltered SW radiance from the total:

lw tot sw Iu = I − Iu , (5)

sw lw 197 The unﬁltered radiances (Iu and Iu ) will be used hereafter to derive the daytime mean

198 radiative ﬂux. Although NISTAR L1B data provide observations every second, hourly data

199 (smoothed with 4-hour running mean) are used to derive ﬂuxes because of the level of noise

200 presented in the measurements (DSCOVR NISTAR data quality report v02).

201 3. Global daytime shortwave and longwave anisotropic

202 factors

203 To derive the global daytime mean SW and LW ﬂuxes from the NISTAR unﬁltered

204 radiances, the anisotropy of the TOA radiance ﬁeld must be considered. The CERES Edition

8 205 4 empirical ADMs and a cloud property composite based upon lower Earth orbit satellite

206 retrievals are used here to estimate the global mean shortwave and longwave anisotropic

207 factors.

208 a. CERES ADMs

209 The Edition 4 CERES ADMs (Su et al. 2015) are constructed using the CERES ob-

210 servations taken during the rotating azimuth plane (RAP) scan mode. In this mode, the

211 instrument scans in elevation as it rotates in azimuth, thus acquiring radiance measurements

212 from a wide range of viewing combinations. The CERES ADMs are derived for various scene

213 types, which are deﬁned using a combination of variables (e.g., surface type, cloud fraction,

214 cloud optical depth, cloud phase, aerosol optical depth, precipitable water, lapse rate, etc).

215 To provide accurate scene type information within CERES footprints, imager (Moderate

216 Resolution Imaging Spectroradiometer (MODIS) on Terra and Aqua) cloud and aerosol re-

217 trievals (Minnis et al. 2010, 2011) are averaged over CERES footprints by accounting for

218 the CERES point spread function (PSF, Smith 1994) and are used for scene type classiﬁca-

219 tion. Over a given scene type (χ), the CERES measured radiances are sorted into discrete

220 angular bins. Averaged radiances (Iˆ) in all angular bins are calculated and all radiances in

221 the upwelling directions are integrated to provide the ADM ﬂux (Fˆ). The ADM anisotropic

222 factors (R) for scene type χ are then calculated as: ˆ ˆ πI(θ0, θ, φ, χ) πI(θ0, θ, φ, χ) R(θ0, θ, φ, χ) = 2π π = , (6) R R 2 ˆ Fˆ(θ , χ) 0 0 I(θ0, θ, φ, χ)cosθsinθdθdφ 0

223 where θ0 is the solar zenith angle, θ is the CERES viewing zenith angle, and φ is the relative

224 azimuth angle between CERES and the solar plane.

225 b. EPIC composite data

226 As stated in the section above, anisotropy of the radiation ﬁeld at the TOA was con-

227 structed for diﬀerent scene types, which were deﬁned using many variables including cloud

9 228 properties such as cloud fraction, cloud optical depth, and cloud phase (Loeb et al. 2005;

229 Su et al. 2015). Although the EPIC L2 cloud product includes threshold-based cloud mask,

230 which identifies the EPIC pixels as high confident clear, low confident clear, high confident

231 cloudy, and low conﬁdent cloudy (Yang et al. 2018), the low resolution of EPIC imagery

2 232 (24×24 km ) and its lack of infrared channels diminish its capability to identify clouds and

233 to accurately retrieve cloud properties. As EPIC lacks the channels that are suitable for

234 cloud size and phase retrievals (Meyer et al. 2016), two cloud optical depths are determined

235 assuming the cloud phase is liquid or ice using constant cloud eﬀective radius (14µm for

236 liquid and 30µm for ice) for cloudy EPIC pixels. These cloud properties are not suﬃcient

237 to provide the scene type information necessary for ADM selections. Therefore, more accu-

238 rate cloud property retrievals are needed to provide anisotropy characterizations to convert

239 radiances to ﬂuxes.

240 To accomplish this, we take advantage of the cloud property retrievals from multiple im-

241 agers on low Earth orbit (LEO) satellites and geostationary (GEO) satellites. The LEO satel-

242 lite imagers include the MODerate-resolution Imaging Spectroradiometer (MODIS) on the

243 Terra and Aqua satellites, the Visible Infrared Imaging Suite(VIIRS) on the Suomi-National

244 Polar-orbiting Partnership satellite, and the Advanced Very High Resolution Radiometer

245 (AVHRR) on the NOAA and MetOps platforms. The GEO imagers are on the Geostation-

246 ary Operational Environmental Satellites (GOES), the Meteosat series, and Himawari-8 to

247 provide semi-global coverage. All cloud properties were determined using a common set of

248 algorithms, the Satellite ClOud and Radiation Property retrieval System (SatCORPS, Min-

249 nis et al. 2008b, 2016), based on the CERES cloud detection and retrieval system (Minnis

250 et al. 2008a, 2010, 2011). Cloud properties from these LEO/GEO imagers are optimally

251 merged together to provide a seamless global composite product at 5-km resolution by us-

252 ing an aggregated rating that considers ﬁve parameters (nominal satellite resolution, pixel

253 time relative to the EPIC observation time, viewing zenith angle, distance from day/night

254 terminator, and sun glint factor to minimize the usage of data taken in the glint region) and

10 255 selects the best observation at the time nearest to the EPIC measurements. About 80% of

256 the LEO/GEO satellite overpass times are within 40 minutes of the EPIC measurements,

257 while 96% are within two hours of the EPIC measurements. Most of the regions covered by

◦ ◦ 258 GEO satellites (between around 50 S and 50 N) have a very small time diﬀerence, in the

259 range of ±30 minutes, because the availability of hourly GEO observations. The polar re-

260 gions are also covered very well by polar orbiters. Thus, larger time diﬀerences are generally

◦ ◦ 261 occurred over the 50 to 70 latitude regions. Given the temporal resolution of the currently

262 available GEO/LEO satellites, this is the best collocation possible for those latitudes.

263 The global composite data are then remapped into the EPIC FOV by convolving the

264 high-resolution cloud properties with the EPIC point spread function (PSF) deﬁned with

265 a half-pixel accuracy to produce the EPIC composite. As the PSF is sampled with half-

266 pixel accuracy, the nominal spacing of the PSF grid is about the same size as in the global

267 composite data. Thus, the accuracy of the cloud fraction in the EPIC composite is not

268 degraded compared to the global composite (Khlopenkov et al. 2017). PSF-weighted averages

269 of radiances and cloud properties are computed separately for each cloud phase, because the

270 LEO/GEO cloud products are retrieved separately for liquid and ice clouds (Minnis et al.

271 2008b). Ancillary data (i.e. surface type, snow and ice map, skin temperature, precipitable

272 water, etc.) needed for anisotropic factor selections are also included in the EPIC composite.

273 These composite images are produced for each observation time of the EPIC instrument

274 (typically 300 to 600 composites per month). Detailed descriptions of the method and the

275 input used to generate the global and EPIC composites are provided in Khlopenkov et al.

276 (2017).

277 Figure 3(a) shows an image from EPIC taken on May 15, 2017 at 12:17 UTC, the cor-

278 responding total cloud fraction (the sum of liquid and ice cloud fractions) from the EPIC

279 composite is shown in 3(b). The liquid and ice cloud fraction, optical depth, and eﬀective

280 height are shown in Figure 3(c-h). For this case, most of the clouds are in the liquid phase.

281 Optically thick liquid clouds with eﬀective heights of 2 to 4 km are observed in the northern

11 282 Atlantic ocean and in the Arctic. Ice clouds with eﬀective heights of 8 to 10 km are observed

283 oﬀ the west coast of Africa and Europe.

284 c. Calculating global daytime anisotropic factors

285 To determine the global daytime mean anisotropic factors, we use the anisotropies char-

286 acterized in the CERES ADMs and they are selected based upon the scene type information

287 provided by the EPIC composite for every EPIC FOV. For a given EPIC FOV (j), its

288 anisotropic factor is determined based upon the Sun-EPIC viewing geometry and the scene

289 identiﬁcation information provided by the EPIC composite:

ˆ e e e e e e πIj(θ0, θ , φ , χ ) Rj(θ0, θ , φ , χ ) = , (7) ˆ e Fj(θ0, χ )

e e 290 where θ is the EPIC viewing zenith angle, φ is the relative azimuth angle between EPIC

e ˆ 291 and the solar plane, and χ is the scene identiﬁcation from the EPIC composite. Here Ij ˆ 292 is the radiance from CERES ADMs and Fj is the ﬂux from CERES ADMs (see Eq. 6).

293 To derive the global mean anisotropic factor, we follow the method developed by Su et al.

294 (2018) and calculate the global daytime mean ADM radiance as:

PN ˆ e e e Ij(θ0, θ , φ , χ ) Iˆ = j=1 . (8) N

295 To calculate the global mean ADM flux, we first grid the ADM flux (Fˆ) for each EPIC

◦ ◦ 296 pixel into 1 latitude by 1 longitude bins (Fˆ(lat, lon)). These gridded ADM ﬂuxes are then

297 weighted by cosine of latitude to provide the global daytime mean ADM ﬂux:

PM Fˆ (lat, lon)cos(lat ) ˆ j=1 j j F = P . (9) cos(latj)

298 The global mean anisotropic factor is calculated as:

πIˆ R = . (10) Fˆ

12 299 We use Rsw and Rlw to denote the mean SW and LW anisotropic factors. The mean SW

300 anisotropic factor is then used to convert the NISTAR SW unﬁltered radiance to ﬂux:

sw sw πIu Fn = . (11) Rsw

301 The LW ﬂux is similarly derived from the following:

lw lw πIu Fn = . (12) Rlw

302 Figure 4 shows an example of SW and LW anisotropic factors for every EPIC FOV. The

303 SW anisotropic factors are generally smaller over clear than over cloudy oceanic regions.

304 Over land, however, the SW anisotropic factors are larger over clear regions than over cloudy

305 regions because of the hot spot eﬀect, which leads to anisotropic factors greater than 1.6

306 over clear land regions at large viewing zenith angles. The LW anisotropic factors show

307 much less variability compared to the SW anisotropic factors, with limb darkening being

308 the dominant feature. The mean SW and LW anisotropic factors for this case are 1.275 and

309 1.041, respectively.

310 4. NISTAR shortwave and longwave ﬂux

311 The temporal resolution of the NISTAR Level 1B data is one second, however, meaning-

312 ful changes in the data only occur over many shutter cycles (each shutter cycle is 4 minutes)

313 due to the demodulation algorithm, which includes a box car ﬁlter having the width of a shut-

314 ter period. The ﬁlter reduces noise and rejects higher harmonics of the shutter frequency.

315 Following demodulation, signiﬁcant instrument noise remains. Therefore, further averag-

316 ing in time over a minimum of 2 hours is recommended to further reduce the noise levels

317 (https://eosweb.larc.nasa.gov/project/dscovr/DSCOVR NISTAR Data Quality Report V02.pdf).

318 In this study, we use hourly radiances averaged from 4-hour running means as suggested by

319 the NISTAR instrument team. The hours that are coincident with the EPIC image times

13 320 are converted to ﬂuxes using the global anisotropic factors calculated using the EPIC com-

321 posites for scene identiﬁcation. Figure 5 shows the hourly SW and LW ﬂuxes derived from

322 NISTAR for April (a) and July (b) 2017. For both months, the SW ﬂuxes ﬂuctuate around

−2 −2 323 210 Wm , with the diﬀerence between daily maximum and minimum as large as 30 Wm .

−2 324 The LW ﬂuxes ﬂuctuate around 260 Wm , and exhibit surprisingly large diurnal variations.

325 These NISTAR ﬂuxes are compared to the CERES Synoptic radiative ﬂuxes and clouds

326 product (SYN1deg, Doelling et al. 2013), which provides hourly cloud properties and ﬂuxes

◦ ◦ 327 for each 1 latitude by 1 longitude. Within the SYN1deg data product, ﬂuxes between

328 CERES observations are inferred from hourly GEO visible and infrared imager measure-

◦ ◦ 329 ments between 60 S and 60 N using observation-based narrowband-to-broadband radiance

330 and radiance-to-ﬂux conversion algorithms. However, the GEO narrowband channels have

331 a greater calibration uncertainty than MODIS and CERES. Several procedures are imple-

332 mented to ensure the consistency between the MODIS-derived and GEO-derived cloud prop-

333 erties, and between the CERES ﬂuxes and the GEO-based ﬂuxes. These include calibrating

334 GEO visible radiances against the well-calibrated MODIS 0.65 µm radiances by ray-matching

335 MODIS and GEO radiances; applying similar cloud retrieval algorithms to derive cloud prop-

336 erties from MODIS and GEO observations; and normalizing GEO-based broadband ﬂuxes

337 to CERES ﬂuxes using coincident measurements. Comparisons with broadband ﬂuxes from

338 Geostationary Earth Radiation Budget (GERB, Harries et al. 2005) indicate that SYN1deg

339 hourly ﬂuxes are able to capture the subtle diurnal ﬂux variations. Comparing with the

−2 340 GERB ﬂuxes, the bias of the SYN SW ﬂuxes is 1.3 Wm , the monthly regional all-sky SW

−2 −2 341 ﬂux RMS error is 3.5 W m , and the daily regional all-sky SW ﬂux RMS error is 7.8 W m

342 (Doelling et al. 2013). These uncertainties could be overestimated, as the GERB domain

343 has a disproportionate number of strong diurnal cycle regions as compared with the globe.

344 To account for the missing energy from the daytime portion that is not observed by the

345 NISTAR (Ah in Figure 1b), and the energy from the nighttime sliver that are within the

346 DSCOVR view (Ad in Figure 1b, only applicable to LW ﬂux), the hourly gridded SYN ﬂuxes

14 347 are integrated by considering only the grid boxes that are visible to NISTAR to produce the

348 global mean daytime ﬂuxes that are comparable to those from the NISTAR measurements: P Fjcos(latj)ωj Fsyn = P . (13) cos(latj)ωj

349 Here Fj is the gridded hourly CERES SYN ﬂuxes, lat is the latitude, and ω indicates whether

350 a grid box is visible to NISTAR (=1 when visible, =0 when not visible). Figure 6a) shows

351 an example of the gridded SYN SW ﬂuxes at 13 UTC on February 1, 2017. SW ﬂuxes for

352 the daytime grid boxes are shown in color, while all nighttime grid boxes are shown in white.

353 Figure 6b) shows the daytime areas (in red) and the nighttime areas (in grey) visible to the

354 NISTAR view. Daytime areas of northern high latitude and North America are not within

355 the NISTAR view and are therefore not included in the comparison with the NISTAR ﬂuxes,

356 and the nighttime slivers in the southern high latitude of Indian Ocean and Paciﬁc Ocean

357 are included in the LW ﬂux comparison with the NISTAR.

358 Figure 7 compares the SW ﬂuxes from NISTAR with those from CERES SYN1deg

359 product integrated for the NISTAR view (Eq. 19) for April (a) and July (b) 2017. The

−2 −2 360 CERES SW ﬂuxes oscillate around 200 Wm and 195 Wm for April and July, whereas

−2 361 the NISTAR counterparts are about 10 to 20 Wm greater. The maxima and minima of

362 SW ﬂuxes from NISTAR align well with those from CERES, though the diﬀerences between

363 daily maximum and minimum from NISTAR appear to be larger than those from CERES.

364 The diurnal variations of SW ﬂux derived from EPIC showed a much better agreement with

365 those from CERES (Su et al. 2018). The exact cause for these larger diurnal variations

366 from NISTAR SW ﬂux is not known. LW ﬂux comparisons are shown in Figure 8. The

−2 367 daily maximum-minimum LW diﬀerences from CERES are typically less than 15 Wm

368 and exhibit small day-to-day and month-to-month variation. However, the daily maximum-

−2 −2 369 minimum LW diﬀerences from NISTAR can vary from 10 Wm to 50 Wm . These larger

370 than expected variability of NISTAR LW ﬂuxes are due to the fact that noise and oﬀset

371 variabilities from both the NISTAR total and SW channel are present in the NISTAR LW

372 radiances. The NISTAR LW ﬂuxes are consistently greater than CERES LW ﬂuxes by about

15 −2 373 10 to 20 Wm in April. The LW ﬂuxes agree better for July, but the NISTAR LW ﬂuxes

374 show larger diurnal variations than the CERES ﬂuxes.

375 Figure 9 compares the SW and LW ﬂuxes from CERES SYN1deg product with those

−2 376 from NISTAR at all coincident hours of 2017. The mean SW ﬂuxes are 203.7 Wm and

−2 −2 377 217.0 Wm , respectively, for CERES and NISTAR, and the RMS error is 14.6 Wm (Fig-

−2 −2 378 ure 9a). The mean LW ﬂuxes are 246.0 Wm and 252.8 Wm for CERES and NISTAR,

−2 379 and the RMS error is 10.5 Wm (Figure 9b). Tables 3 and 4 summarize the ﬂux com-

380 parisons between NISTAR and CERES for all months of 2017. The NISTAR SW ﬂuxes

381 are consistently greater than those from CERES SYN1deg by about 3.4% to 7.8%, and the

382 NISTAR LW ﬂuxes are also greater than those from CERES SYN1deg by 1.0% to 5.0%.

383 Furthermore, the SW ﬂuxes from NISTAR are highly correlated (correlation coeﬃcient of

384 about 0.89) with those from CERES SYN1deg, but the correlation for the LW ﬂuxes are

385 rather low (correlation coeﬃcient is about 0.38).

386 NISTAR ﬂuxes derived at the EPIC image times are averaged into daily means and are

387 compared with the daily means from CERES SYN1deg using concurrent hours (Figure 10).

388 The NISTAR SW ﬂuxes are consistently higher than those from CERES by about 10 to 15

−2 389 Wm . CERES SW ﬂuxes show a strong annual cycle, which is driven by the incident solar

390 radiation that is aﬀected by the Earth-Sun distance. This annual cycle is also evident in the

391 NISTAR SW fluxes, albeit the fluxes during the period from April to August are flatter than

392 those from CERES. The NISTAR LW ﬂuxes are greater than those from CERES except

−2 393 during the boreal summer months, with the largest diﬀerence of 10 Wm in February and

−2 394 the smallest diﬀerence of a few Wm during the boreal summer months. The CERES LW

−2 395 ﬂuxes show an annual cycle of about 10 Wm , with the largest LW ﬂuxes occurring during

396 the boreal summer when the vast land masses of the northern hemisphere are warmer than

397 during the other seasons. The annual cycle of the NISTAR LW ﬂuxes shows less seasonal

−2 398 variation. From April to October, the NISTAR LW ﬂuxes oscillate around 255 Wm , and

−2 399 oscillate around 250 Wm for other months. Additionally, the CERES LW ﬂuxes exhibit

16 400 much smaller day-to-day variations than their NISTAR counterparts. Note some of the

401 variations of daily mean ﬂuxes shown in Figure 10 are due to temporal sampling changes

402 when data transmissions encountered diﬃculties and/or during spacecraft maneuvers.

403 5. Conclusions and discussions

404 The SW radiances included in the NISTAR L1B data are ﬁltered radiances and the eﬀect

405 of the ﬁlter transmission must be addressed before these measurements can be used to derive

406 any meaningful ﬂuxes. A comprehensive spectral radiance database has been developed

407 to investigate the relationship between ﬁltered and unﬁltered radiances using theoretically

408 derived values simulated for typical Earth scenes and the NISTAR spectral transmission

409 functions. The ratio between ﬁltered and unﬁltered SW radiances is very stable, varying

410 less than 0.3% for the scenes and the Sun-viewing geometries included in the database. The

411 mean ratio of 0.8690 is used to derive the unﬁltered SW radiance from the NISTAR L1B

412 ﬁltered SW radiance measurements.

413 To convert these unfiltered radiances into fluxes, the anisotropy of the radiance field must

414 be taken into account. We use the scene-type dependent CERES angular distribution models

415 to characterize the global SW and LW anisotropy. These global anisotropies are calculated

416 based upon the anisotropies for each EPIC pixel. To accurately account for the anisotropy for

417 each EPIC pixel, an EPIC composite was developed which includes all information needed

418 for angular distribution model selections. The EPIC composite includes cloud property

419 retrievals from multiple imagers on LEO and GEO satellites. Cloud properties from these

420 LEO and GEO imagers are optimally merged together to provide a global composite product

421 at 5-km resolution by using an aggregated rating that considers several factors and selects the

422 best observation at the time nearest to the EPIC measurements. The global composite data

423 are then remapped into the EPIC FOV by convolving the high-resolution cloud properties

424 with the EPIC PSF to produce the EPIC composite. PSF-weighted averages of radiances

17 425 and cloud properties are computed separately for each cloud phase, and ancillary data needed

426 for anisotropic factor selections are also included in the EPIC composite.

427 These global anisotropies are applied to the NISTAR radiances to produce the global

◦ 428 daytime SW and LW ﬂuxes and they are validated against the CERES Synoptic 1 latitude

◦ 429 by 1 longitude ﬂux product. Only the grid boxes that are visible to the NISTAR view

430 are integrated to produce the global mean daytime ﬂuxes that are comparable to the ﬂuxes

431 from the NISTAR measurements. The NISTAR SW ﬂuxes are consistently greater than

−2 −2 432 those from CERES SYN1deg by 10 Wm to 15 Wm (3.3% to 7.8%), but these two SW

433 ﬂux datasets are highly correlated indicating that the diurnal and seasonal variations of

434 the SW ﬂuxes are fairly similar for both of them. The NISTAR LW ﬂuxes are also greater

435 than those from CERES SYN1deg, but the magnitude of the diﬀerence has larger month-

−2 436 to-month variations than that for the SW ﬂuxes. The largest diﬀerence of about 14 Wm

−2 437 (∼5.5%) occurred in April 2017 and the smallest diﬀerence of about ∼4 Wm (∼1.6%)

438 occurred during July. Furthermore, the NISTAR LW ﬂuxes have very low correlations with

439 the CERES LW fluxes. NISTAR LW fluxes exhibit a nearly flat annual variation, whereas

440 the CERES LW ﬂuxes exhibit a distinct annual cycle with the highest LW ﬂux occurs in

441 July when the vast northern hemisphere land masses are warmest. The NISTAR LW ﬂuxes

442 also exhibit unrealistically large day-to-day variations.

443 The SW ﬂux discrepancy between NISTAR and CERES is caused by: 1) CERES instru-

444 ment calibration uncertainty, 2) CERES ﬂux algorithm uncertainty, 3) NISTAR instrument

445 measurement uncertainty, and 4) NISTAR ﬂux algorithm uncertainty. The CERES SW

446 channel calibration uncertainty is 1% (1σ, McCarthy et al. 2011; Priestley et al. 2011; Loeb

−2 447 et al. 2018), which corresponds to about 2.1 Wm for daytime mean SW ﬂuxes. The

−2 448 CERES algorithm uncertainty includes radiance-to-ﬂux conversion error, which is 1.0 Wm

449 according to Su et al. (2015), and diurnal correction uncertainty, which is estimated to be

−2 450 1.9 Wm when Terra and Aqua are combined (Loeb et al. 2018). The NISTAR SW channel

−2 451 measurement uncertainty is 2.1%, which corresponds to 4.4 Wm . The NISTAR algorithm

18 452 uncertainty is essentially the radiance-to-ﬂux conversion error. The estimation of this error

453 source is not readily available given the unique NISTAR viewing perspective. However, if

454 we assume the discrepancy between EPIC derived SW ﬂux and CERES SW ﬂux (Su et al.

455 2018) is also from uncertainty sources 1) and 2) listed above, plus the EPIC calibration,

456 narrowband-to-broadband conversion, and radiance-to-ﬂux conversion for EPIC, then we

457 can deduce that the radiance-to-ﬂux conversion uncertainty for the NISTAR viewing geom-

−2 458 etry should be less than 2 Wm . Thus the total diﬀerence expected from these uncertainty

2 2 2 2 2 1/2 −2 459 sources should be (2.1 + 1.9 + 1.0 + 4.4 + 2.0 ) = 5.7 Wm .

460 Similarly, the LW ﬂux discrepancy between NISTAR and CERES is due to the same

−2 461 sources of error. The daytime CERES LW ﬂux uncertainty from calibration is 2.5 Wm

462 (1σ, Loeb et al. 2009). The CERES LW radiance-to-ﬂux conversion error is about 0.75

−2 −2 463 Wm (Su et al. 2015), and diurnal correction uncertainty is estimated to be 2.2 Wm

464 (Loeb et al. 2018). However, the CERES LW ADMs were developed without taking the

465 relative azimuth angle into consideration, which has little impact on the CERES LW ﬂux

466 accuracy because of its Sun-synchronous orbit. Given that the NISTAR only views the Earth

467 from the backscattering angles, the LW ﬂux uncertainty due to radiance-to-ﬂux conversion

468 could be larger for the clear-sky footprints (Minnis et al. 2004). As the clear-sky occurrences

469 are small at the EPIC footprint size level, our best estimate of this uncertainty is no more

−2 470 than 0.4 Wm . The calibration uncertainty for NISTAR LW is deduced from the calibration

471 uncertainties of total and SW channels. The total channel calibration uncertainty is 1.5%,

−2 −2 472 which is about 6.8 Wm assuming the total radiative energy of 450 Wm . The SW channel

−2 473 measurement uncertainty is 4.4 Wm . The resulting LW channel measurement uncertainty

2 2 1/2 −2 474 is thus equal to (6.8 + 4.4 ) = 8.1 Wm . Although no direct estimation of the radiance-

475 to-ﬂux conversion uncertainty for LW is available, we do not expect that it exceeds its SW

−2 476 counterpart of 2.0 Wm . Thus the total diﬀerence expected from these uncertainty sources

2 2 2 2 2 2 1/2 −2 477 should be (2.5 + 0.75 + 0.4 + 2.2 + 8.1 + 2.0 ) = 9.1Wm .

478 The uncertainty sources listed above can explain part of the SW ﬂux diﬀerences and

19 479 all of the LW ﬂux diﬀerences between CERES and NISTAR. The error sources related to

480 NISTAR are preliminary and are under careful evaluation. Although the LW ﬂux diﬀerences

481 between CERES and NISTAR are within the uncertainty estimation, the correlation between

482 NISTAR and CERES is rather low, about 0.38. This is because the NISTAR LW radiance

483 is derived as the diﬀerence between total channel radiance and SW channel radiance, thus

484 noise and oﬀset variability of both the NISTAR total and SW channels are present in the

485 NISTAR LW ﬂuxes. As a result, more variability is expected in the LW data which leads to

486 the low correlation. Although the noise level present in the NISTAR measurements prevent

487 the production of high frequency SW ﬂux, the current 4-hour running mean ﬂuxes are highly

488 correlated with the CERES product. The NISTAR SW ﬂux can be used to test the diurnal

489 variations of SW ﬂux in the high-temporal resolution model outputs from the Coupled Model

490 Intercomparison Project. Furthermore, the spectral ratio information from NISTAR presents

491 a new way to evaluate the models and opens a new perspective on exoplanet observations

492 (Carlson et al. 2019).

493 Acknowledgments.

494 This research was supported by the NASA DSCOVR project. The CERES data were

495 obtained from the NASA Langley Atmospheric Science Data Center at

496 https://eosweb.larc.nasa.gov/project/ceres/ssf terra-fm1 ed4a table(ssf aqua-fm3 ed4a table).

497 The data used to produce the ﬁgures and tables in this paper are available to readers upon

498 request. We thank the DSCOVR project managed by Richard Eckman for support.

20 499

500 REFERENCES

501 Carlson, B. E., A. A. Lacis, C. Colose, A. Marshak, W. Su, and S. Lorentz, 2019: Spectral

502 signature of the biosphere: NISTAR ﬁnds it in our solar system from the Lagrangia L-1

503 point. Geophys. Res. Lett., doi:10.1029/2019GL083736.

504 Doelling, D. R., N. G. Loeb, D. F. Keyes, M. L. Nordeen, D. Morstad, B. A. Wielicki, D. F.

505 Young, and M. Sun, 2013: Geostationary enhanced temporal interpolation for CERES ﬂux

506 products. J. Atmos. Oceanic Technol., 30, 1072–1090, doi:10.1175/JTECH-D-12-00136.1.

507 Harries, J. E., J. E. Russell, J. A. Hanaﬁn, H. Brindley, J. Futyan, J. Rufus, and coauthours,

508 2005: The geostationary earth radiation budget project. Bull. Am. Meteor. Soc., 86, 945–

509 960.

510 House, F. B., A. Gruber, G. E. Hunt, and A. T. Mecherikunnel, 1986: History of satellite

511 missions and measurements of the Earth radiation budget (1957-1984). Rev. Geophys., 24,

512 357–377.

513 Kato, S., N. G. Loeb, and K. Rutledge, 2002: Estimate of top-of-atmosphere albedo for a

514 molecular atmosphere over ocean using Clouds and the Earth’s Radiant Energy System

515 measurements. J. Geophys. Res., 107 (D19), doi:10.1029/2001JD001309.

516 Kato, S., et al., 2011: Improvements of top-of-atmosphere and surface irradiance computa-

517 tion with CALIPSO-, and MODIS-derived cloud and aerosol properties. J. Geophys. Res.,

518 116 (D19209), D19 209, doi:10.1029/2011JD016050.

519 Khlopenkov, K., D. Duda, M. Thieman, P. Minnis, W. Su, and K. Bedka, 2017: Devel-

520 opment of multi-sensor global cloud and radiance composites for Earth radiation budget

21 521 monitoring from DSCOVR. Remote sensing of clouds and the atmosphere XXII, A. Com-

522 eron, E. I. Kassianov, K. Schafer, R. H. Picard, and K. Weber, Eds., Warsaw, Poland,

523 Proc. SPIE 10424, Vol. 10424K (2 October 2017), doi:10.1117/12.2278645.

524 Loeb, N. G., S. Kato, K. Loukachine, and N. Manalo-Smith, 2005: Angular distribution

525 models for top-of-atmosphere radiative ﬂux estimation from the clouds and the earth’s

526 radiant energy system instrument on the terra satellite. part I: Methodology. J. Atmos.

527 Oceanic Technol., 22, 338–351.

528 Loeb, N. G., J. M. Lyman, G. C. Johnson, R. P. Allan, D. R. Doelling, T. Wong, B. J.

529 Soden, and G. L. Stephens, 2012: Observed changes in top-of-the-atmosphere radiation

530 and upper-ocean heating consistent within uncertainty. Nature Geosci., 5, 110–113, doi:

531 10.1038/NGEO1375.

532 Loeb, N. G. and N. Manalo-Smith, 2005: Top-of-atmosphere direct radiative eﬀect of aerosols

533 over global oceans from merged CERES and MODIS observations. J. Climate, 18, 3506–

534 3526.

535 Loeb, N. G., N. Manalo-Smith, W. Su, M. Shankar, and S. Thomas, 2016: CERES top-of-

536 atmosphere Earth radiation budget climate data record: Accounting for in-orbit changes

537 in instrument calibration. Remote Sens., 8 (182), doi:10.3390/rs8030182.

538 Loeb, N. G., K. J. Priestley, D. P. Kratz, E. B. Geier, R. N. Green, B. A. Wielicki, P. O.

539 Hinton, and S. K. Nolan, 2001: Determination of unﬁltered radiances from the Clouds

540 and the Earth’s Radiant Energy System instrument. J. Appl. Meteor., 40, 822–835.

541 Loeb, N. G. and G. L. Schuster, 2008: An observational study of the relationship between

542 cloud, aerosol and meteorology in broken low-level cloud conditions. J. Geophys. Res.,

543 113 (D14214), D14 214, doi:10.1029/2007JD009763.

544 Loeb, N. G., B. A. Wielicki, D. R. Doelling, G. L. Smith, D. F. Keyes, S. Kato, N. Manalo-

22 545 Smith, and T. Wong, 2009: Towards optimal closure of the Earth’s top-of-atmosphere

546 radiation budget. J. Climate, 22, 748–766, doi:10.1175/2008JCLI2637.1.

547 Loeb, N. G., et al., 2018: Clouds and the Earth’s Radiant Energy System (CERES) Energy

548 Balanced and Filled (EBAF) Top-of-Atmosphere (TOA) Edition-4.0 Data Product. J.

549 Climate, 31, 895–918, doi:10.1175/JCLI-D-17-0208.1.

550 McCarthy, J. M., H. Bitting, T. A. Evert, M. E. Frink, T. R. Hedman, P. Skaguchi, and

551 M. folkman, 2011: A summary of the performance and long-term stability of the pre-

552 launch radiometric calibration facility for the Clouds and the Earth’s Radiant Energy

553 System (CERES) instruments. 2011 IEEE International Geoscience and Remote Sensing

554 Symposium, 1009–1012, doi:10.1109/IGARSS.2011.6049304.

555 Meyer, K., Y. Yang, and S. Platnick, 2016: Uncertainties in cloud phase and optical thickness

556 retrievals from the Earth Polychromatic Imaging Camera (EPIC). Atmos. Meas. Tech., 9,

557 1785–1797, doi:10.5194/amt-9-1785-2016.

558 Minnis, P., K. Bedka, Q. Z. Trepte, C. R. Yost, S. T. Bedka, B. Scarino, K. V. Khlopenkov,

559 and M. M. Khaiyer, 2016: A consistent long-term cloud and clear-sky radiation property

560 dataset from the Advanced Very High Resolution Radiometer (AVHRR). Climate Algo-

561 rithm Theoretical Basis Document (C-ATBD). CDRP-ATBD-0826 Rev 1–NASA,NOAA

562 CDR Program, doi:10.7289/V5HT2M8T.

563 Minnis, P., A. V. Gambheer, and D. R. Doelling, 2004: Azimuthal anisotropy of longwave and

564 infrared window radiances from the Clouds and the Earth’s Radiant Energy System on the

565 Tropical Rainfall Measuring Mission on Terra satellites. J. Geophys. Res., 109 (D08202),

566 doi:10.1029/2003JD004471.

567 Minnis, P., et al., 2008a: Cloud detection in nonpolar regions for CERES using TRMM VIRS

568 and TERRA and AQUA MODIS data. IEEE Trans. Geosci. Remote Sensing, 46 (11),

569 3857–3884.

23 570 Minnis, P., et al., 2008b: Near-real time cloud retrievals from operational and research

571 meteorological satellites. Proc. SPIE 7108, Remote Sens. Clouds Atmos. XIII, Cardiﬀ,

572 Wales, UK, doi:10.1117/12.800344.

573 Minnis, P., et al., 2010: CERES Edition 3 cloud retrievals. 13th Conference on Atmospheric

574 Radiation, Am. Meteorol. Soc., Oregon, Portland.

575 Minnis, P., et al., 2011: CERES Edition-2 cloud property retrievals using TRMM VIRS

576 and TERRA and AQUA MODIS data, Part I: Algorithms. IEEE Trans. Geosci. Remote

577 Sensing, 49 (11), 4374–4400, doi:10.1109/TGRS.2011.2144601.

578 Pincus, R., C. P. Batstone, R. J. P. Hofmann, K. E. Taylor, and P. J. Glecker, 2008: Evalu-

579 ating the present-day simulation of clouds, precipitation, and radiation in climate models.

580 J. Geophys. Res., 113 (D14209), D14 209, doi:10.1029/2007JD009334.

581 Priestley, K. J., et al., 2011: Radiometric performance of the CERES Earth radiation budget

582 climate record sensors on the EOS Aqua and Terra spacecraft through April 2007. J.

583 Atmos. Oceanic Technol., 28, 3–21, doi:10.1175/2010JTECHA1521.1.

584 Quaas, J., O. Boucher, N. Bellouin, and S. Kinne, 2008: Satellite-based estimate of the

585 direct and indirect aerosol climate forcing. J. Geophys. Res., 113 (D05204), D05 204,

586 doi:10.1029/2007JD008962.

587 Satheesh, S. K. and V. Ramanathan, 2000: Large diﬀerences in tropcial aerosol forcing at

588 the top of the atmosphere and earth’s surface. Nature, 405, 60–63.

589 Smith, G. L., 1994: Eﬀects of time response on the point spread function of a scanning

590 radiometer. Appl. Opt., 33, 7031–7037.

591 Stephens, G. L., et al., 2012: An update on Earth’s energy balance in light of the latest

592 global observations. Nature Geosci., 5, 691–696, doi:10:1038/NGEO1580.

24 593 Su, W., A. Bodas-Salcedo, K.-M. Xu, and T. P. Charlock, 2010a: Comparison of the trop-

594 ical radiative flux and cloud radiative effect profiles in a climate model with Clouds and

595 the Earth’s Radiant Energy System (CERES) data. J. Geophys. Res., 115 (D01105),

596 D01 105, doi:10.1029/2009JD012490.

597 Su, W., J. Corbett, Z. A. Eitzen, and L. Liang, 2015: Next-generation angular distribution

598 models for top-of-atmosphere radiative ﬂux calculation from the CERES instruments:

599 Methodology. Atmos. Meas. Tech., 8, 611–632, doi:10.5194/amt-8-611-2015.

600 Su, W., N. G. Loeb, G. L. Schuster, M. Chin, and F. G. Rose, 2013: Global all-sky shortwave

601 direct radiative forcing of anthropogenic aerosols from combined satellite observations and

602 GOCART simulations. J. Geophys. Res., 118, 1–15, doi:10.1029/2012JD018294.

603 Su, W., N. G. Loeb, K. Xu, G. L. Schuster, and Z. A. Eitzen, 2010b: An estimate of aerosol

604 indirect eﬀect from satellite measurements with concurrent meteorological analysis. J.

605 Geophys. Res., 115 (D18219), D18 219, doi:10.1029/2010JD013948.

606 Su, W., et al., 2018: Determining the shortwave radiative ﬂux from Earth Polychromatic

607 Imaging Camera. J. Geophys. Res., 123, doi:10.1029/2018JD029390.

608 Trenberth, K. E., J. T. Fasullo, and J. Kiehl, 2009: Earth’s global energy budget. Bull. Am.

609 Meteor. Soc., 90, 311–323, doi:10.1175/2008BAMS2634.1.

610 Wang, H. and W. Su, 2013: Evaluating and understanding top of the atmosphere cloud

611 radiative eﬀects in Intergovernmental Panel on Climate Change (IPCC) ﬁfth assessment

612 report (AR5) cloupled model intercomparison project phase 5 (CMIP5) models using

613 satellite observations. J. Geophys. Res., 118, 1–17, doi:10.1029/2012JD018619.

614 Wielicki, B. A., B. R. Barkstrom, E. F. Harrison, R. B. Lee, G. L. Smith, and J. E. Cooper,

615 1996: Clouds and the Earth’s Radiant Energy System (CERES): An Earth Observing

616 System experiment. Bull. Amer. Meteor. Soc., 77, 853–868.

25 617 Wild, M., D. Folini, C. Schar, N. G. Loeb, E. G. Dutton, and G. Konig-Langlo, 2013:

618 The global energy balance from a surface perspective. Clim. Dyn., 40, 3107–3134, doi:

619 10.1007/s00382-012-1569-8.

620 Yang, Y., et al., 2018: Cloud Products from the Earth Polychromatic Imaging Camera

621 (EPIC): Algorithms and Initial Evaluaiton. Atmos. Meas. Tech. Discussions, 1–23, doi:

622 10.5194/amt-2018-316.

623 Zhang, J., S. A. Christopher, L. A. Remer, and Y. J. Kaufman, 2005: Shortwave aerosol

624 radiative forcing over cloud-free oceans from Terra: 2. Seasonal and global distributions.

625 J. Geophys. Res., 110 (D10S24), D10S24, doi:10.1029/2004JD005009.

26 626 List of Tables

627 1 Summary of the cases included in the spectral radiance database. AOD is for

628 aerosol optical depth, COD is for cloud optical depth. 28

629 2 Mean ratio and standard deviation (in parenthesis) of ﬁltered radiance to

630 unﬁltered radiance for SW and NIR bands over diﬀerent scene types. 29

631 3 SW ﬂux comparisons between NISTAR and CERES SYN1deg for all coinci-

−2 632 dent observations of 2017. Fn is the NISTAR ﬂux (in Wm ), Fs is the SYN

−2 633 ﬂux (in Wm ), and the root mean square (RMS) error between them (in

634 Wm−2). 30

635 4 LW ﬂux comparisons between NISTAR and CERES SYN1deg for all coinci-

−2 636 dent observations of 2017. Fn is the NISTAR ﬂux (in Wm ), Fs is the SYN

−2 637 ﬂux (in Wm ), and the root mean square (RMS) error between them (in

638 Wm−2). 31

27 Table 1. Summary of the cases included in the spectral radiance database. AOD is for aerosol optical depth, COD is for cloud optical depth.

Clear AOD Aerosol type Surface Ocean 8 6 4 Land 8 4 15 Snow 5 2 5 Cloudy COD Cloud type Surface Atmosphere Ocean 7 4 liquid and 3 ice 4 4 Land 7 4 liquid and 3 ice 15 1

28 Table 2. Mean ratio and standard deviation (in parenthesis) of filtered radiance to unfiltered radiance for SW and NIR bands over different scene types.

SW ratio (standard deviation × 1000) 0.0 29.0 41.4 60.0 75.5 85.0 Clear Ocean 0.8659(1.0) 0.8660(1.0) 0.8661(1.1) 0.8664(1.2) 0.8669(1.0) 0.8674(0.8) Clear Land 0.8694(0.6) 0.8693(0.6) 0.8692(0.6) 0.8690(0.5) 0.8687(0.5) 0.8685(0.8) Clear Snow 0.8689(0.1) 0.8689(0.1) 0.8689(0.2) 0.8688(0.2) 0.8688(0.3) 0.8687(0.4) Cld Ocean 0.8687(1.0) 0.8687(1.0) 0.8688(0.9) 0.8688(0.8) 0.8688(0.7) 0.8687(0.6) Cld Land 0.8694(0.4) 0.8693(0.3) 0.8693(0.3) 0.8692(0.3) 0.8690(0.4) 0.8689(0.5) NIR ratio (standard deviation × 1000) 0.0 29.0 41.4 60.0 75.5 85.0 Clear Ocean 0.8293(23.1) 0.8270(24.0) 0.8253(25.5) 0.8235(28.3) 0.8238(28.4) 0.8229(26.4) Clear Land 0.8790(9.6) 0.8777(10.4) 0.8764(10.7) 0.8730(10.8) 0.8663(10.1) 0.8501(12.4) Clear Snow 0.8360(1.7) 0.8360(1.8) 0.8361(1.9) 0.8363(2.1) 0.8370(2.8) 0.8365(6.0) Cld Ocean 0.8557(3.2) 0.8555(2.6) 0.8562(2.4) 0.8567(3.1) 0.8565(4.4) 0.8539(7.9) Cld Land 0.8627(8.2) 0.8624(7.8) 0.8621(7.3) 0.8613(6.2) 0.8598(4.8) 0.8566(6.2)

29 Table 3. SW flux comparisons between NISTAR and CERES SYN1deg for all coincident −2 −2 observations of 2017. Fn is the NISTAR flux (in Wm ), Fs is the SYN flux (in Wm ), and the root mean square (RMS) error between them (in Wm−2).

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Fs — 208.1 203.4 199.8 201.0 200.2 194.4 193.0 198.7 208..9 221.6 228.2 Fn — 218.5 215.4 211.5 214.1 213.5 209.2 208.7 211.2 222.8 235.1 240.0 RMS — 11.9 14.0 12.9 14.0 14.6 16.0 16.8 13.9 15.5 14.5 14.0

30 Table 4. LW flux comparisons between NISTAR and CERES SYN1deg for all coincident −2 −2 observations of 2017. Fn is the NISTAR flux (in Wm ), Fs is the SYN flux (in Wm ), and the root mean square (RMS) error between them (in Wm−2).

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Fs — 242.0 241.1 243.0 246.3 249.1 251.5 248.9 245.5 242.9 239.8 240.6 Fn — 253.1 248.1 257.7 255.8 255.2 255.6 253.2 255.5 253.5 250.4 253.3 RMS — 13.4 10.0 16.0 11.5 10.3 8.7 10.0 12.2 12.5 12.4 14.4

31 639 List of Figures

640 1 Schematic of a) Earth-Sun-DSCOVR geometry and b) Earth disc that are

641 visible to the L1 DSCOVR view (left with an area fraction of At) and to the

642 L2 view (right). The golden area on the left shows the daytime area fraction

643 (Av) that are visible to DSCOVR, the black area on the left shows the night

644 portion (Ad) that are within the DSCOVR view, and the golden area on the

645 right is the daytime portion (Ah) missed by the DSCOVR. Not to scale. 34

646 2 NISTAR SW and NIR spectral transmission function. 35

647 3 EPIC RGB image for May 15, 2017 at 12:17 UTC (a), and the corresponding

648 total cloud fraction (b, in %). Liquid and ice cloud fractions are shown in

649 (c) and (d), liquid and ice cloud optical depths are shown in (e) and (f), and

650 liquid and ice cloud eﬀective height (in km) are shown in (g) and (h). (b) to

651 (h) are all derived from the EPIC composite. 36

652 4 SW anisotropic factors (a) and LW anisotropic factors (b) derived from the

653 CERES ADMs using the EPIC composite for scene identiﬁcation for May 15,

654 2017 at 12:17 UTC. 37

655 5 SW ﬂux (blue) and LW ﬂux (red) derived from NISTAR measurements for

656 April (a) and July (b), 2017. 38

657 6 An example of the daytime SW ﬂux distributions from CERES SYN1deg

658 product at 13 UTC on February 1, 2017 (a), and the corresponding daytime

659 areas (in red) and nighttime areas (in grey) that are visible to NISTAR and

660 the terminator boundary (in blue) (b). 39

−2 661 7 SW ﬂux (in Wm ) comparisons between NISTAR and CERES SYN for April

662 (a) and July (b) 2017. 40

−2 663 8 LW ﬂux (in Wm ) comparisons between NISTAR and CERES SYN for April

664 (a) and July (b) 2017. 41

32 665 9 Comparison of coincident hourly SW and LW ﬂuxes from NISTAR and CERES

666 SYN1deg for 2017. Color bar indicates the number of occurrence. 42

667 10 Daily mean SW ﬂux (a) and LW ﬂux (b) comparisons between CERES SYN1deg

668 (blue) and NISTAR (red) for 2017. 43

33 Ad Av Ah

Fig. 1. Schematic of a) Earth-Sun-DSCOVR geometry and b) Earth disc that are visible to the L1 DSCOVR view (left with an area fraction of At) and to the L2 view (right). The golden area on the left shows the daytime area fraction (Av) that are visible to DSCOVR, the black area on the left shows the night portion (Ad) that are within the DSCOVR view, and the golden area on the right is the daytime portion (Ah) missed by the DSCOVR. Not to scale.

34 Fig. 2. NISTAR SW and NIR spectral transmission function.

35 Fig. 3. EPIC RGB image for May 15, 2017 at 12:17 UTC (a), and the corresponding total cloud fraction (b, in %). Liquid and ice cloud fractions are shown in (c) and (d), liquid and ice cloud optical depths are shown in (e) and (f), and liquid and ice cloud eﬀective height (in km) are shown in (g) and (h). (b) to (h) are all derived from the EPIC composite.

36 Fig. 4. SW anisotropic factors (a) and LW anisotropic factors (b) derived from the CERES ADMs using the EPIC composite for scene identiﬁcation for May 15, 2017 at 12:17 UTC.

37 Fig. 5. SW ﬂux (blue) and LW ﬂux (red) derived from NISTAR measurements for April (a) and July (b), 2017.

38 Fig. 6. An example of the daytime SW ﬂux distributions from CERES SYN1deg product at 13 UTC on February 1, 2017 (a), and the corresponding daytime areas (in red) and nighttime areas (in grey) that are visible to NISTAR and the terminator boundary (in blue) (b).

39 Fig. 7. SW ﬂux (in Wm−2) comparisons between NISTAR and CERES SYN for April (a) and July (b) 2017.

40 Fig. 8. LW ﬂux (in Wm−2) comparisons between NISTAR and CERES SYN for April (a) and July (b) 2017.

41 Fig. 9. Comparison of coincident hourly SW and LW ﬂuxes from NISTAR and CERES SYN1deg for 2017. Color bar indicates the number of occurrence.

42 Fig. 10. Daily mean SW ﬂux (a) and LW ﬂux (b) comparisons between CERES SYN1deg (blue) and NISTAR (red) for 2017.

43 Interactive comment on “Determining the Daytime Earth Radiative Flux from National Institute of Standards and Technology Advanced Radiometer (NISTAR) Measurements” by Wenying Su et al. Anonymous Referee #1 Received and published: 18 August 2019

11. In ERB calibration your definition of filtered radiance as IRRADIANCE/ SOLIDANGLE is only true if the instrument has a completely flat spectral response, which from fig 2 is certainly not the case for NISTAR SW & NIR channels.

We are not sure what the reviewer means here. NISTAR is a broadband instrument, it measures the energy from the spectral ranges defined in Fig. 2. The relationship between radiance and irradiance should not change with the spectral response function.

113-132. This is based on CERES unfiltering of Loeb et al 2001 I assume (where they are also labelled Eqns 3 & 4 at https://journals.ametsoc.org/doi/pdf/10.1175/1520- 0450%282001%29040%3C0822%3ADOURFT%3E2.0.CO%3B2 ). Is it completely identical to CERES using the same decades old CERES radiative transfer database? This might be important to briefly mention as it could help eliminate mere inversion biases when you compare to CERES later in the paper.

The concept of unfiltering used here is the same as that by Loeb et al. (2001), but the database used here is different and was calculated specifically for this study. The current database contains 722 clear-sky cases and 1519 cloudy-sky cases (line 166), whereas the total number of cases used by Loeb et al. (2001) was 272. This information is added to the manuscript.

142. This is confusing, although I accept it probably amounts to same thing, are look up tables of a & b values (Eqns 3 & 4) or a table of kappa ratios actually used? Only if both techniques are used separately should there be 4 rather than just 2 eqns?

Thank you for catching this. The Equations 3 and 4 are the original method used to unfilter the CERES observations. As you know, we have scene-type information and Sun-viewing geometry for each CERES footprint, thus the regression can be applied based upon the scene type and Sun-viewing geometry of the CERES footprint. NISTAR views the entire Earth as a single pixel, and the cloud fraction, cloud type, and land/ocean portions differ from time to time. Luckily, the NISTAR SW spectral response function is such that the ratio between filtered and unfiltered radiances exhibit very little sensitivity to the scene types and Sun-viewing geometry. We rewrote the section on page 7 and 8 to correct this.

143. How are spectrally dependent changes to the transmission of the quartz filter due to outgassing contamination measured after launch and throughout the mission?

On-orbit measurements indicated that the filters have not degraded significantly since they were measured on the ground during calibration. On orbit measurements of the broadband transmission of the filter stack are continually made every three months using the earth as a source and the photodiode as a detector. The ratios of the on-orbit transmittances amongst each of the two sets of 3 nominally identical filters of each type (SW and NIR) are within 0.2% of each other (as expected from ground measurements) and have remained stable to less than 0.1% throughout the mission. In the case of the SW filter (quartz), the on-orbit broadband transmittance is within 1% of the spectral transmittance of the filter stack over the wavelength range from 500 nm to 2500 nm.

144. What about quartz filter leakage? Are you using the NIR channel for that somehow (similar to Loeb et al 2001 above)?

A thin quartz filter can transmit significantly at wavelengths greater than many tens of micrometers, however, the NISTAR filter stacks consists of a pair of 3 mm thick quartz substrates—one is a bare uncoated substrate and the other has dielectric coatings to block light below about 700nm. At 3 mm thickness per substrate the transmittance below 100 micrometers is negligible. Loeb et al (2001) did not use any NIR channel.

146. Are you sure no unfiltering of the total channel is required, if so how? Was its spectral response measured to be certain? How are you certain no changes to the effective gains of the cavity channels due to electronics radiation exposure are occurring? I’m assuming you do not have onboard blackbodies?

We do not unfilter the total channel. The total channel spectral response is determined by the spectral absorptance of its cavity absorber, which, like a blackbody, relies on multiple reflections to achieve a high degree of absorptance (emissivity). Each cavity is conical in shape to trap light and is painted with a specular black paint, Z302, which has a very small component of diffuse reflectance. Measurements of the cavity absorptance made on the ground at wavelengths of 488 nm, 514 nm and 632 nm confirmed that the cavity absorbed more than 0.9997 of the incident light. Given the known spectral reflectance of Z302 to long wavelengths and the cavity design (verified at visible wavelengths), un-filtering of the total channel is not required.

The only electronics that affect the cavity channel gains are those that measure the electric power applied to the cavity heaters. Those electronics were chosen for their radiation tolerance and long term stability. Given the on-orbit radiation exposure levels, such degradation is not expected to significantly affect heater power measurements. Similar techniques and electronics are used to measure the total solar irradiance from space with a stability of less than 0.1%, which is sufficient to resolve the 11 year solar cycle. Unlike those measurements, degradation from UV exposure is not an issue here. You are correct, there aren’t any on-board blackbodies to use as a references. Such blackbodies would have to have phase transition temperature references to be less sensitive to radiation exposure than the radiometers. This is because electronic temperature measurements are much more challenging than measurement of the power applied to the cavity heaters.

147. How are the SW and Total channels balanced in the solar region as in Kratz et al 2002? (https://agupubs.onlinelibrary.wiley.com/doi/epdf/10.1029/2001JD001170)

Since the NISTAR instrument only views the sunlit side of the Earth, there are no measurements taking during nighttime that can be used in the same manner as Kratz et. al. (2002), in which they looked at the correlation between nighttime total channel and window channel.

261. what is a shutter cycle? Why is a boxcar filter used in the demodulation algorithm? Are details of these processes important?

NISTAR utilizes a shutter to modulate light from the Earth just as a chopper wheel is used in the laboratory to modulate a light source. The shutter is opened and closed continuously with a 50% duty cycle with a period of nominally 4 minutes. Each 4 minute period is a shutter cycle. The demodulation algorithm is analogous to what is performed in a digital lock-in amplifier. Use of a boxcar filter having the width of a shutter period strongly rejects higher harmonics of the shutter frequency. Other low pass filters could be used. Note that additional filtering at lower frequencies, e.g., 4 hour running averages, are used to further reduce noise levels. Description is added on page 5.

264. Recommended based on what, the URL does not work?

Based on the noise level. The URL was temporally unavailable due to internal web maintenance, it should be available now. Sorry about that.

266. Why 4 hour “running means” and how is this different from a 4-hour wide boxcar filter from the terminology you used earlier? Does this mean a 4-hour running mean is taken of the boxcar filtered then 2-hour averaged data? Why are the 4 hour means suggested by the NISTAR instrument team?

A running mean of 4 hours is conceptually the same as a boxcar filter. The 4 hour averages are additional filtering that occurs after the 4 minute wide boxcar filter to reduce noise levels. A four hour compromise is proposed as a trade-off between reducing noise and attenuating the signal of interest, however, the data is also provided without the additional filtering so the user may apply their own filter.

286. These GERB comparisons need a reference.

Reference "Doelling et al. (2013)" was provided on line 342, immediately after summarizing the comparison results.

311. Why does the onboard data processing cause this?

We removed this sentence in the revised version.

315. How are the offsets countered, space looks?

Yes. The shutter removes some, but not all offsets. Those that remain are removed with monthly space looks. Description is added on page 5.

332. With as few as 10 EPIC results per day are these always equally spaced in time? If not, could this not lead to biases?

When EPIC is in normal operations, it receives about 10 images daily during the winter cadence. They are normally spaced about 2 hours apart. EPIC receives about 20 images a day during the summer cadence and they are about 1 hour apart. If we simply compare the daily mean fluxes averaged using the EPIC image times with those averaged over the 24 hours, that would lead to biases. In this study, we only averaged the CERES SYN1deg using the hours that coincide with the EPIC times (line 376). Thus ensure both daily means are calculated using same number of hours.

338. So it seems the LW difference is greatest in Northern Hemisphere Winter, when more ocean is observed? This may be a calibration artifact or error in knowledge of the NISTAR SW channel for the UV region. As per the point above for line 147, how are you balancing SW and Total channels to assure accurate LW in daylight?

Preliminary analysis of the 2018 measurements does not show the same difference pattern (i.e. larger difference over the boreal summer months than the winter months), thus not supporting the hypothesis of the reviewer. As we mentioned earlier, NISTAR only views the sunlit side of the Earth and the same method used by Kratz et al (2002) cannot be applied here.

352. “A comprehensive spectral database has been developed”, so is it different from that used by CERES?

Yes, and more details are added on page 7-8.

355. So is a constant of 0.8690 used for all NISTAR unfiltering? Unfiltering of LEO scenes varies greatly by several percent especially for ocean scenes etc. So, it seems a value of 0.3% difference for primarily land vs Pacific Ocean scenes would vary more (and maybe adds to your seasonal cycle). What results lead to the 0.3% conclusion and did you try a scene by scene unfiltering?

Based on the simulated filtered and unfiltered radiances for 722 clear-sky cases and 1519 cloudy-sky cases for each Sun-viewing geometry, the ratio between filtered and unfiltered radiances is extremely stable (see Table 2). Table 2 summarized the ratios and their standard deviation for each solar zenith angle bin for each scene type. For clear-sky case, each solar zenith angle bin contains over 57,000 simulations; and for cloudy-sky case, each solar zenith angle bin contains over 120,000 simulations. The largest ratio difference over different scene types happens under overhead sun, where the ratio for clear ocean is 0.8659 and is 0.8694 for clear land. Using constant unfiltering ratio of 0.8690, it could cause up to 0.3% unfiltering uncertainty if a clear ocean scene is encountered. However, NISTAR views the sunlit side of the Earth as a single pixel. There are always clouds and land mixed in. Thus we state the unfiltering uncertainty should be less than 0.3%. We rewrote the unfiltering portion of the paper on page 7 and 8 to clarify the reviewer's concerns.

370. Is this the PSF of the EPIC telescope separate from its array of detectors? How was it measured?

We are not sure we understand the reviewer's question. The PSF tells us where does the light measured in one pixel come from. It’s a function of the instrument’s entire optical system, telescope and detector. The EPIC PSF was measured in the laboratory before launch, nominal PSF is given in Khlopenkov et al. (2017, SPIE).

388. Again, this could be due to a constant unfiltering factor?

Please see response above.

392. Loeb et al 2018 only quotes the 1% accuracy figure as do you, please provide a peer reviewed SI traceable reference.

The following CERES calibration references are added:

J. M. McCarthy, H. Bitting, T. A. Evert, M. E. Frink, T. R. Hedman, P. Skaguchi, and M. folkman. A summary of the performance and long-term stability of the pre-launch radiometric calibration facility for the Clouds and the Earth’s Radiant Energy System (CERES) instruments. In 2011 IEEE International Geoscience and Remote Sensing Symposium, pages 1009–1012, 2011.

K. J. Priestley, G. L. Smith, S. Thomas, D. Cooper, R. B. Lee, D. Walikainen, P. Hess, Z. P. Szewczyk, and R. Wilson. Radiometric performance of the CERES Earth radiation budget climate record sensors on the EOS Aqua and Terra spacecraft through April 2007. J. Atmos. Oceanic Technol., 28:3–21, 2011.

396. Please give a peer reviewed reference for the 2.1% NISTAR SW accuracy figure.

NISTAR is a relatively new instrument and so far no peer reviewed publication describing the calibration is available. The presentation describing the NISTAR calibration is available at: https://avdc.gsfc.nasa.gov/pub/DSCOVR/Science_Team_Meeting_Sept_2019/L1/NISTAR_Godda rd%20Science%20Team%2020190917.pdf

404. With so many error sources not well known it is wrong to simply add them all in quadrature, which assumes they are all random and independent. A more sophisticated error analysis is needed.

The reviewer is correct that the error sources considered here were simply added to approximate the uncertainty. I would say this is a simplified estimate of the uncertainty, but not the wrong estimate. We know the sources of the uncertainty, but don't know the correlation of all the error sources and therefore unable to estimate the covariances of the sources considered here. The uncertainty given here can be regarded as the upper bound, and this method has been used by Loeb et al. (2009) and Loeb et al. (2018).

407. The 1.8Wmˆ2 accuracy for CERES LW applies for nighttime LW only. During the day which is always the case for NISTAR it is less accurate. This is because it requires the earlier discussed balancing of the SW and Total channel which if done wrong can result in measuring the Earth warmer at night than during the day for example (see Fig11b, Page 14 at https://journals.ametsoc.org/doi/pdf/10.1175/2010JTECHA1521.1). Hence for NISTAR which only views day LW, this is an important consideration.

The reviewer is correct that the accuracy of the daytime and nighttime LW is different. The daytime LW uncertainty due to calibration is 2.5 Wm-2 (1 sigma). The combined uncertainty is updated based on the daytime LW flux uncertainty (line 461).

415. Guesstimate? This is most unsatisfactory for any science paper, let alone one on climate measurements. Please do better.

Changed to "estimate".

423. Again, adding in quadrature for so many uncertain, often modelling terms is not acceptable. For example, consider how the error in knowledge of SW vs Total solar response could be systematic because of an error in the ground lab, it will partly cancel in the Total – SW subtraction.

Please see our response above regarding the uncertainty estimation. The daytime LW flux uncertainty due to calibration is estimated by accounting for the calibration uncertainty in both total channel and SW channel, and the correlations between these two channels.

428. This is true, in addition to the above-mentioned systematic nature of SW and Total errors not considered in your quadrature additions. A more sophisticated analysis is needed.

As we stated above, the error analysis considered both SW and total channel. However, changes in error analysis won't affect the correlation between the LW flux from CERES and from NISTAR.

Overall this paper has merit but needs work to fill in the blanks on some of the processes/references used. The large differences of NISTAR from CERES appears strange and would seem at first look to be largely from algorithm errors. I feel this could be acceptable being a new measurement, but needs to be stated more clearly in the paper as such.

The NISTAR instrument is the first ever cavity radiometer placed at the L-1 point to measure the Earth's radiation. EPIC on board the DSCOVR also provides 10 narrowband observations from the same Sun-viewing geometry and the visible channels of EPIC are calibration against MODIS. When the global SW anisotropic factors were applied to the EPIC broadband radiance (derived by applying narrowband-to-broadband regressions to EPIC blue, green, and red measurements), the EPIC SW flux agrees with the CERES SYN SW flux to within 2%. The good agreement indicates that the algorithm that we developed is accurate and is not the cause for the large discrepancy between NISTAR and CERES SYN. Even though there are discrepancies between the NISTAR fluxes and CERES SYN fluxes, we feel it is important to document the measurement, the algorithm, and the validation for future reference.

The use of constant SW unfiltering also raises concern and leads to the possibility it is a cause of the larger than expected seasonal cycles, but more investigation is needed. Also some insight in the introduction into the purpose of NISTAR would be good, such as giving illustration if and how it complements the climate observing system discussed by Weilicki et al 2013 (https://journals.ametsoc.org/doi/pdf/10.1175/BAMS-D-12-00149.1 ). In summary, this paper could become suitable for publication, given more work, research and additions that address the points above. It should then be re-considered under peer review.

Interactive comment on “Determining the Daytime Earth Radiative Flux from National Institute of Standards and Technology Advanced Radiometer (NISTAR) Measurements” by Wenying Su et al. Anonymous Referee #3 Received and published: 19 August 2019

General comments: This paper presents the scheme and algorithm of deriving TOA SW/LW flux from NISTAR measurements and comparison also made with the corresponding results derived from CERES. I am impressed by the detailed and clear description of the algorithms. The paper is very well written and relevant to the community. I recommend publication after addressing the minor issues listed bellowed. It doesn’t seem that the uncertainties in the algorithms would give a consistent bias seeing in the differences between NISTAR and CERES. Has there been analysis with the NISTAR instrument measurements and calibration? The low correlation between NISTAR LW flux and that of CERES is puzzling. To bypass the potential uncertainties in part of the algorithms, it may be useful to look at the correlation between the NISTAR LW radiances and the CERES flux to see if they are correlated at all.

The NISTAR instrument team (who produces the L1 data) is responsible for the instrument calibration and the team has presented their calibration at the DSCOVR science team meetings(https://avdc.gsfc.nasa.gov/pub/DSCOVR/Science_Team_Meeting_Sept_2019/L1/NIST AR_Goddard%20Science%20Team%2020190917.pdf). So far their analysis are mainly focused on the SW channel. NISTAR has three broadband electrical substitution radiometers (ESRs). All ESRs have a large background noise as they measure the change in incident optical power. Two steps are utilized to remove the background noise: first using a shutter to modulate the source which removes most of the background noise then using dark space view to remove the residual shutter-modulated background. The shutter modulated background is largest for the total channel and is smaller for the SW channel. As the LW is derived from the difference between total and SW channels, both total channel and SW channel background noises contribute to the LW uncertainty. The NISTAR total channel uncertainty is 1.5% and the SW channel uncertainty is 2.1%. Assuming the SW flux is 210 Wm-2 and the LW flux is 240 Wm-2, thus gives the total flux uncertainty as 450*1.5%=6.8Wm-2, and the SW flux uncertainty as 210*2.1%=4.4Wm-2. The resulted uncertainty in LW flux is 8.1 Wm-2, which can explain most of the LW differences between NISTAR and CERES SYN shown in Table 4. See added description on page 6. The low correlation is also caused by the background noise in both the total and SW channels. Details on NISTAR calibration are added on pages 5 and 6. Below is an example of August 2017, where the correlation between CERES SYN LW and NISTAR LW flux is about 0.38, and the correlation between CERES SYN LW flux and the NISTAR LW radiance is about 0.41. It is obvious that the low correlation is mainly from the instrument calibration.

Specific comments: Page 5 and 6: The authors have derived the regression equations for the unfiltered radiances (Eq 3 and 4); what is the reason for using the less accurate ratio method (Eq. 5 and 6)?

The Equations 3 and 4 are the original method we planned to use for the NISTAR unfiltering. But unlike other LEO instruments that have scene-type information and Sun-viewing geometry for each footprint, and the regression can be applied based upon the scene type and Sun-viewing geometry of each footprint. NISTAR views the entire Earth as a single pixel, and the cloud fraction, cloud type, and land/ocean portions differ from time to time. Luckily, the NISTAR SW spectral response function is such that the ratio between filtered and unfiltered radiances exhibit very little sensitivity to the scene types and Sun-viewing geometry. We rewrote the sections on page 7 and 8 to correct this.

Page 14: How are the portion of the Earth not visible to NISTAR decided? Also, similar to NISTAR missing some of the daytime portion of the Earth, it must be seeing part of the night time side of the Earth. Are these taking into account for the longwave calculations?

The mask is calculated based upon the solar zenith angle and the EPIC viewing zenith angle and each EPIC pixel is identified as nighttime hidden to EPIC, or nighttime visible to EPIC, or daytime hidden to EPIC, or daytime visible to EPIC. Both the daytime and nighttime visible to EPIC are considered for the CERES SYN product to compare with the NISTAR LW measurements. Some clarification is added on page 14 and Figure 6b) is modified accordingly.

Review on “Determining the Daytime Earth Radiative Flux from National Institute of Standards and Technology Advanced Radiometer (NISTAR) Measurements” by Su et al.

This paper documents the methodology to derive the broadband radiative flux from the measurements of the NISTAR instrument onboard of the DSCOVR mission. Some preliminary results based on this method are compared with the well-developed CERES data. The SW fluxes derived from the NISTAR compares reasonably well with CERES, but the LW fluxes from NISTAR have a systematic bias and low correlation coefficient when benchmarked with CERES.

The topic of this paper is important and suitable for AMT. The paper is well organized. However, the paper lacks some important technical details about the instrument and the methodology, as well as the author’s opinion about the usefulness of the NISTAR product. In my view, some significant revisions are needed before the paper can be accepted for publication. Below is a list of questions and concerns I have.

1) The parameterization scheme described in Section 2 to obtain unfiltered radiance from observed filtered radiance is confusing. Up to line 132, the method seems to be based on the polynomial parameterization scheme in Eqs (3) and (4). But then it suddenly changed to the simply ratio-based parameterization in Eqs. (5) and (6). Why are there two types of parameterization? Which one is used?

2) What is the FOV size of the NISTAR instrument? Does it observe the earth pixel by pixel (similar to EPIC) or as a whole? Does its FOV include some cosmic background and, if so, how is that treated?

NISTAR observes the entire sunlit side of the Earth as one pixel. We specifically mentioned this on lines 53-54.

3) Within its FOV, does the NISTAR instrument response to the radiance from different locations and angles equally? In other words, do the radiances from the edge of the earth disc have the same weighting as those from the center of the disc?

Yes, NISTAR response to the radiance from different locations and angles equally. Optically the instrument is very simple—there aren’t any lenses or mirrors, just filters, and a pair of apertures, and incident light is nearly perpendicular to the filters and apertures. The Earth subtends an angle of less than 1 degree from DSCOVR.

4) It is stated that “The biases in the anisotropy correction for the DSCOVR scattering angle are mitigated and potentially minimized by the wide range of different scene 71 types viewed in a given NISTAR measurement.” Some references are needed to support it.

We referenced the Su et al. (2018) paper here. As this is a very new way to measure the Earth radiative flux, no other references are available.

5) In Su et al. (2018), a similar method is used to derive the fluxes from EPIC measurements. One of the byproduct from this EPIC-based method is the “global day-time mean SW radiance” ��̅.̅̅̅ Is it something directly comparable to the observation of NISTAR instrument? If so, some comparisons should be made because both EPIC and NISTAR have the similar sun-satellite geometry.

Indeed, we have derived the "global daytime mean SW radiances from EPIC". They are consistently lower than the radiances from NISTAR. Below are the comparison between NISTAR and EPIC radiances for April and July 2017. The mean differences are between 4 to 6 Wm-2sr-1. We chose not to include these results in this paper to avoid any confusions and the EPIC and CERES comparisons were provided in Su et al. (2018).

6) I have several questions about the method described in Section 3c. First of all, what is the theoretical based for Eqs 9~ 11? If my understanding is correct, the global mean SW flux is

[ ( ) ( ) ( ) ] � � � , � � , ∅ � , �(�) � = � � �(�, � , ∅ , �)� �

Where r denotes a point on earth. But this is not equal to ∬ [ (),(),∅(),()] . ∬ (, ,∅ ,)

More detailed mathematically derivations are needed here. Secondly, one might ask if a global mean anisotropic factor is even physically meaningful? The average is over a large range of viewing angles and scene types. Does the result have any physical meaning? Moreover, are the angular and spectral averaging independent and can be treated independently? The derivations in Section 3c seem to suggest they are independent, but this is not obvious to me. Some clarification is needed.

We agree with the reviewer that the above two equations are not the same. The first equation is how we calculate global mean flux from low-Earth orbit satellites (i.e. CERES) using the footprint level data (resolution on the order of 20 km) by first grid the data then area weight to calculate the global mean. We did not use the second equation in our study to derive the fluxes from NISTAR radiance measurements.

To derive the global mean flux from NISTAR measurements, a corresponding anisotropic factor to characterize the sunlit portion of the Earth as a whole is needed, and this is the definition of the global mean anisotropic factor we used in the paper. The global mean anisotropic factor is derived by using the radiances and fluxes defined in the CERES angular distribution models (ADMs). The global mean radiance and flux from CERES ADMs were calculated independently (see Equations 8 and 9 in the revised version). They are used to derive the global anisotropic factor (Equation 10) and subsequently to convert the NISTAR radiance to flux (Equation 11). The deviation of the NISTAR flux used here is not the same as illustrated by the second equation above. This method has been tested for both the NISTAR and EPIC measurements.

7) This paper only shows “how to do it” but does not explain “why to do it” other than it can be done. I understand that this paper is to document the method used to derive the flux from the radiance observations of NISTAR. But I think in addition to the technical details the reader would apricate some insights and opinions from the authors about the usefulness of the product. We already have the state-of-the-art CERES flux product and in Su et al. (2018) flux product has also been developed. What is new/novel/important about the NISTAR flux product other than the fact it can be done? What kind of applications can this product be used for? Some discussions about these important questions should be added to the abstract and conclusion parts.

We added some information on NISTAR measurement and its utility in the introduction (lines 59-68). We also added some perspective on the utility of NISTAR SW fluxes in the conclusion section (lines 486-492). Review of “Determining the Daytime Earth Radiative Flux from National Institute of Standards and Technology Advanced Radiometer (NISTAR) Measurements” by Su et al. 2018 General comments: This manuscript derives sunlit side of the Earth’s radiation budget (SW and LW) from a single pixel measurement of NISTAR instrument on board the DSCOVR mission and compares with the radiation fluxes derived from the CERES measurements. This is a very interesting and important work as the Earth’s radiation budget has been so far solely measured by the ERBE/CERES project and there are very little independent and direct measurements of these important quantities. This work builds upon many previous works the team has been working on for many years including narrowband broadband conversion, ADM, GEO/LEO composite cloud products etc. The paper is well written and structured. I do have some questions and suggestions regarding the derivation of global ADM and evaluation of each components of the fluxes.

Specific comments: Line 98: What is the uncertainty level of NISTAR L1B radiance? What kind of calibration procedures have been used to produce the L1B radiance? You have discussed some of the issues later in the paper but it’s worthwhile to have a paragraph to discuss the NISTAR at the beginning of the paper. NISTAR provides a completely different methodology of estimating the earth’s radiation budget and independent check of Earth’s radiation budget created from CERES measurements, the difference found in this article is very serious and should be adequately explained. NISTAR’s absolute calibration and uncertainty is of fundamental importance, otherwise the readers would question the well-established CERES products.

The NISTAR instrument team (who produce the L1 data) is responsible for the instrument calibration and the team has presented their calibration at the DSCOVR science team meetings (https://avdc.gsfc.nasa.gov/pub/DSCOVR/Science_Team_Meeting_Sept_2019/L1/NISTAR_Godd ard%20Science%20Team%2020190917.pdf). So far their analysis are mainly focused on the SW channel. NISTAR has three broadband electrical substitution radiometers (ESRs). All ESRs have a large background noise as they measure the change in incident optical power. Two steps are utilized to remove the background noise: first using a shutter to modulate the source which removes most of the background noise then using dark space view to remove the residual shutter-modulated background. The shutter modulated background is largest for the total channel and is smaller for the SW channel. As the LW is derived from the difference between total and SW channels, both total channel and SW channel background noises contribute to the LW uncertainty. The NISTAR total channel uncertainty is 1.5% and the SW channel uncertainty is 2.1%. More details on NISTAR calibration is added on page 4-6.

Line 147. The conversion from filtered to unfiltered radiances used the ratio derived from model simulation data using eq 5 and 6. Why not using the regression (3) and (4)? The regression indicates the ratio could not be constant because it’s a quadratic function and has an offset. It’s justified to use a constant ratio between the two if the ratio varies little as for the SW band, but a constant ratio for NIR would introduce an unnecessary source of error (1_2%) for the NIR and I don’t see why you should abandon the regression.

The Equations 3 and 4 are the original method we planned to use for the NISTAR unfiltering. But unlike other LEO instruments that have scene-type information and Sun-viewing geometry for each footprint, and the regression can be applied based upon the scene type and Sun-viewing geometry of each footprint. NISTAR views the entire Earth as a single pixel, and the cloud fraction, cloud type, and land/ocean portions differ from time to time. Luckily, the NISTAR SW spectral response function is such that the ratio between filtered and unfiltered radiances exhibit very little sensitivity to the scene types and Sun-viewing geometry. As we don't have the scene- type information, the regression method can't be applied to NIR either. We rewrote the sections on pages 7 and 8 to correct this.

Line 152: Did you use NIR in this work? If not, could you explain why NISTAR takes the NIR measurement?

We did not use the NIR channel in our work. When NISTAR was design in the 1990s, the primary utility for the NIR channel is to study the enhanced absorption of SW radiation by clouds (Collins 1998) and more recently by Carlson et al. (2019) to look at the spectral ratio of the sunlit side of the Earth and the potential of using the ratio for model evaluation (see introduction on page 3).

Line 187: EPIC images have 8x8 km2 resolution at nadir and are 1/cos(vza) larger at larger view zenith angles. The EPIC cloud products are retrieved at its native resolution with (2014x2014) pixels in a granule. Some channels have degraded into 1024x1024 for downlink but reversed to 2014x2014 afterwards. Equation (9) and (11), Ij and Fj seem to refer to radiance and flux in each EPIC composite pixel. Do you actually use those in the mean ADM calculations? If yes, did you use the EPIC measured narrowband radiances to compute the broadband radiance and flux for each pixel?

In Equations (9) to (11) (now Equations 7-9), the ^I and ^F refer to the radiances and fluxes from the CERES ADMs (same symbols are used in Equation 6). To clarify the confusion, we added a sentence on page 12 (lines 291-292). They are not from EPIC.

Why did you grid the fluxes into 1x1 grid boxes and not the radiances? The global mean flux is computed from Eq. 11 to take care of different sizes of grids in each latitude. If you grid the radiance, then you would compute the mean radiance the same fashion as the flux. Otherwise, if you average the radiance from each pixel directly, then you would also have to consider the pixel size differences and the radiance average has to be a pixel-size weighted average.

Radiance and flux are fundamentally different physical quantities. Radiance is the total amount of energy confined to a given direction per unit surface area (in Wm-2sr-1). One essential property of radiance is that it is additive, meaning if several sources contribute to the radiance at a particular point and in a particular direction, the total radiance is the sum of the radiances from each source as if it were acting alone (Bohren and Clothiaux, 2006). On the other hand, flux is the energy per unit surface area (Wm-2) and need to be area-weighted when compute global means.

If my understanding is correct, then the global ADM not only rely on composite product’s scene identification, CERES ADM for each pixel, but also on EPIC’s radiances measurements (which rely on CERES-MODIS collocation and narrowband to broadband conversion) to derive the global mean ADM. The EPIC-based sunlit global SW flux (Su et al. 2018) has used EPIC radiances and CERES ADMs and does not really need global ADM and thus global ADM is essentially untested. From EPIC radiance to flux, it relies on CERES derived narrowband-broadband conversion and CERES ADM, therefore the EPIC global flux provides some consistent check but not absolute validation in my opinion.

The global ADM is derived using the scene identifications (surface type, cloud fraction, cloud optical depth, etc. ) from EPIC composite to select the anisotropic factors from the CERES ADMs. EPIC radiances are not used.

In Su et al. (2018), we used the same methodology to derive the global SW anisotropy factors (same as Equations 7-10 in the revised version). They are then applied to the EPIC global daytime mean "broadband" SW radiances, which were derived by using narrowband-to- broadband regressions. The EPIC global daytime mean "broadband" SW radiances are analogous to the NISTAR SW measurements, and the same SW anisotropy factors were applied to both NISTAR and EPIC SW radiances to derive SW fluxes. As demonstrated in Su et al. (2018), the SW flux from EPIC agree with CERES SYN to within 2%, which means that the method that we developed to derived the global mean anisotropic factors are robust.

Eq. 13 and 14. From these equations, we know that the NISTAR flux depends on unfiltered radiances from NISTAR and the global ADM derived from EPIC (which itself depend on many other instruments and procedures). I would strongly suggest the authors examine the global ADM and NISTAR’s radiance measurements separately to understand the variability and trends from each of these components. The computation of global ADM can be refined as mean radiance could be computed with pixel-size weighted average. The NISTAR total radiance and NIR radiance are also worth looking at especially when LW is derived from total subtract the SW.

Again, EPIC composite provides the cloud properties that needed to select the anisotropy factors. We don't use any EPIC measurements in this study. The global mean anisotropy factors are calculated by deriving the anisotropic factors for each EPIC pixel. We did examine the NISTAR radiance against the "global daytime mean SW radiances from EPIC", derived by using narrowband-to-broadband regressions (Su et al. 2018). The NISTAR radiances are consistently greater than the EPIC "broadband SW" radiances. Below are the comparison between NISTAR and EPIC radiances for April and July 2017. The mean differences are between 4 to 6 Wm-2sr-1. We chose not to include these results in this paper to avoid any confusions and the EPIC and CERES comparisons were provided in Su et al. (2018).

As noted by the reviewer, the LW radiance is derived by subtracting SW from the total. Thus the LW contains information of the total channel. As the focus of this paper is to derive SW and LW fluxes from NISTAR and validate the product with CERES product, and there aren't any global daytime total and NIR measurements that can be used to compare with the NISTAR measurements, simply looking at the NISTAR total and NIR channel measurement won't add value to the paper.